MCP in practice

Every directory ranks. Almost none publish the shape of what they are ranking. So here is ours, measured across the 343 npm packages that……

For two weeks our package table had a defect we did not notice: it was sorted by downloads, and the top entry was Playwright……

When we review a server we record who runs it. Of 127 reviewed entries, 71 are published by the vendor whose product they connect……

We record the license of every package we compile. The distribution is unsurprising and worth stating plainly: 245 of 343 are MIT (71%), 51……

The naive way to build an MCP directory is to query npm for the keyword and publish what comes back. That produces a directory……

Directories rank. They rarely tell you what the population looks like. So we measured ours: every npm package we could verify declares an MCP……

The first time an MCP server fails, it looks like the model is being stupid. It calls a tool, gets nothing back, and apologizes.……

People budget for conversation tokens and forget the ones that are always there: the system prompt, the tool schemas, and the resource descriptions. They……

Every week someone installs an MCP server from a list, wires it into their agent, and then spends an afternoon debugging a transport mismatch……